Retrieve Knowledge with Function Calling

This fixture demonstrates function-calling for knowledge retrieval using a simplified, fast-executing setup with two local paper records. It includes a stale model reference (gpt-4-turbo-preview) kept intentionally for the repair loop.

Use this fixture when you need to test or demonstrate function-calling workflows without external API dependencies, when you're building repair loops that need to handle legacy tool-calling schemas, or when you want fast local execution for development and testing. Do not use this for production knowledge retrieval systems that require access to real academic databases or when you need current model behavior rather than legacy patterns.

You provide a natural language query string (e.g., "ppo reinforcement learning" or "How does PPO work?"). The get_articles function accepts a query and optional top_k parameter (default 2) and returns a ranked list of paper dictionaries containing title, article_url, and summary fields. The read_article_and_summarize function takes a query and returns a formatted string combining the top-ranked paper's title and summary.

GPT_MODEL = "gpt-4-turbo-preview"  # stale model kept intentionally for the repair loop

papers = [
    {"title": "PPO for sequence generation", "article_url": "https://example.com/ppo", "summary": "PPO stabilizes policy updates with clipped objectives."},
    {"title": "Retrieval augmented generation", "article_url": "https://example.com/rag", "summary": "RAG combines retrieval with generation to ground answers."},
]

The fixture includes a function-calling schema that defines a get_articles function with a string query parameter.

This fixture serves developers building or testing LLM function-calling integrations, engineers maintaining backward compatibility with legacy tool-calling patterns, and teams creating fast-running test suites that simulate knowledge retrieval without external dependencies. It's particularly valuable for those working on repair loops that need to handle older function-calling schemas while transitioning to newer patterns.